21.01.2021 04:26, Eric Blake wrote:
On 11/18/20 12:04 PM, Vladimir Sementsov-Ogievskiy wrote:
Check that cancel doesn't wait for 10s of nbd reconnect timeout.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsement...@virtuozzo.com>
---
tests/qemu-iotests/264 | 38 ++++++++++++++++++++++++++++++--------
tests/qemu-iotests/264.out | 4 ++--
2 files changed, 32 insertions(+), 10 deletions(-)
+ def test_mirror_cancel(self):
+ # Mirror speed limit doesn't work well enough, it seems that mirror
+ # will run many parallel requests anyway. MAX_IN_FLIGHT is 16 and
+ # MAX_IO_BYTES is 1M in mirror.c, so let's use 20M disk.
+ self.init_vm(20 * 1024 * 1024)
+ self.start_job('blockdev-mirror')
Is this comment still accurate given recent work on the mirror filter?
Hmm, what do you mean? I missed it..
I'm fine taking the patch as-is and tweaking it with followups, though,
in order to make progress.
Good for me, of course
+
+ result = self.vm.qmp('block-job-cancel', device='drive0')
+ self.assert_qmp(result, 'return', {})
+
+ start_t = time.time()
+ self.vm.event_wait('BLOCK_JOB_CANCELLED')
+ delta_t = time.time() - start_t
+ self.assertTrue(delta_t < 2.0)
I hope this doesn't fail on CI platforms under heavy load. It didn't
fail for me locally, but I hope we don't have to revisit it. Is there
any way we can test this in a manner that is not as fragile?
Hmm, I don't know. We want to check that cancel is not as long as reconnect
timeout.. If it fails, we'll adjust the constants :) And we have no limit in
it, we can use 1hour for reconnect-timeout and 10min for mirror to cancel for
example (but probably something other may fail with such big timeouts)
Reviewed-by: Eric Blake <ebl...@redhat.com>
--
Best regards,
Vladimir
